Fig. 2: Parsimonious signature diagnostic performance for prevalent TB and signature score distributions in people without HIV.

Representative (a) box-and-whisker plot and (b) receiver operating characteristic (ROC) curve of the parsimonious signature with the best diagnostic performance (Roe3 signature) in the CORTIS-01 study of people without HIV. Box-and-whisker plots and ROC curves for the other signatures are in the Supplementary Information (Fig. S3). The box-and-whisker plot depicts Roe3 signature score (measured at enrolment) distribution by symptom status (each dot represents a participant) in asymptomatic (n = 2570) and symptomatic (n = 83) participants with no TB, participants who progressed to incident TB (n = 29), and participants with prevalent subclinical (asymptomatic; n = 52) and clinical (symptomatic; n = 8) TB. Prevalent and incident TB comprised all primary endpoint cases. Symptoms were recorded at the time of enrolment for participants without TB and those with prevalent TB. p-values for comparison of median signature scores between groups in the box-and-whisker plot were calculated with the Mann–Whitney U test and corrected for multiple comparisons by use of the Benjamini–Hochberg Procedure³⁹. Boxes depict the IQR, the midline represents the median, and the whiskers indicate the IQR ± (1.5 × IQR). The ROC curve depicts diagnostic performance (area under the curve, AUC, with 95% CI) of the Roe3 signature for the primary endpoint (1’ Endpoint), i.e. TB diagnosed on two or more liquid culture-positive or Xpert MTB/RIF-positive sputum samples. The ROC curve shows participants with symptomatic clinical prevalent TB versus symptomatic controls (Symptom+), and participants with asymptomatic, subclinical prevalent TB versus asymptomatic controls (Symptom–). The shaded areas represent 95% CIs. The solid box depicts the optimal criteria (95% sensitivity and 80% specificity) and the dashed box depicts the minimal criteria (90% sensitivity and 70% specificity) set out in the WHO Target Product Profile for a triage test²⁰. c Summary of signature diagnostic performance in the order of primary endpoint AUC estimates. The diagnostic AUC estimates in symptomatic and asymptomatic participant sub-groups are also shown. The midline indicates the AUC estimate, the error bars indicate the 95% CIs, and the black dotted line indicates the lower bound of the 95% CI for the best performing signature for the primary endpoint.